GTD algorithm implementation

[Amir-19]

I was reading the GTD implementation and I got a little confused about updating the weights. Based on the code these are the update rules.

        self.w += alpha*(delta*self.e + gm_p*(1-lm_p)*np.dot(self.e, self.h)*xp)
        self.h += beta*(delta*self.e + np.dot(self.h, x)*x)

However, based on Maei's thesis the update rules are a little different.

        self.w += alpha*(delta*self.e - gm_p*(1-lm_p)*np.dot(self.e, self.h)*xp)
        self.h += beta*(delta*self.e - np.dot(self.h, x)*x)

Is there a reason for these differences?

[rldotai]

That looks like an error on my part.
I referred to code that I'd already written when I was benchmarking these algorithms (for example) but presumably made a typo.

I'll fix that and add some tests to make sure everything else works.
Thanks for point that out!